55-85% of long Claude session usage was re-reading history — dev open-sources /flight-deck fix

PawelHuryn · x · 2026-11-21

Pawel Huryn found that 55-85% of his long agent sessions' Claude usage went to the agent re-reading its own history — every turn re-reads the whole conversation, and 2x length costs 4x. His MIT-licensed /flight-deck skill starts each build phase in a fresh session and keeps state in a STATUS.md file (what's running, blocked, next), rewritable and readable by any tool. He avoids built-in compaction because it's locked inside one session in the provider's format; his builds span Claude Code, Codex, and fresh review sessions.

Original post →

More from coding & agent

coding & agent channel →